Coffee and cancer

(1)

Faculty of Health Sciences, Department of Community Medicine

Coffee and cancer

The Norwegian Women and Cancer Study / Northern Sweden Health and Disease Study

Marko Lukic

A dissertation for the degree of Philosophiae Doctor – June 2018

(2)

Coffee and cancer

The Norwegian Women and Cancer Study / Northern Sweden Health and Disease Study

Marko Lukic

A dissertation for the degree of Philosophiae Doctor (PhD)

Department of Community Medicine Faculty of Health Sciences UiT The Arctic University of Norway

Tromsø, Norway 2018

(3)

Acknowledgments

This work was conducted at the Department of Community Medicine (Institutt for samfunnsmedisin), Faculty of Health Sciences, UiT The arctic University of Norway, from the autumn 2014 to the summer 2018. The work was sponsored by a university grant.

I would like to thank my supervisor Tonje Braaten for sharing your experience and ideas. I cannot thank you enough for meddling into my work only when I asked you to. If anyone, it was you who made me an independent researcher. As that is one of the primary goals of any PhD project – you did an outstanding job. Sorry for not finishing the simulation study. I will, promise…

Eiliv, thank you for telling me three things. First, that it was a really important finding that consuming lot of coffee does not affect the risk of cancer in Norwegian women. I did not care about p values ever since. Second, for telling me that subgroup analyses are “devil’s playground”. Finally, that I should not put too much effort into writing biological explanations of our findings in a discussion!

Idlir, I miss you so much, bro… That should tell you enough.

Bente, Karina, Katja, Kristin, Mie, Therese, Tor Gisle, Torkjel, I feel privileged for having you as friends and colleagues for the past for years. You made my work and life in Norway so much easier.

I would like to thank all of my co-authors, especially to Guri and Elisabete, for their contribution and insightful suggestions and inputs.

I would also like to thank my fellow PhD students. The discussions we had during my PhD were helpful, to say at least.

Women of NOWAC, thank you for participating in the study, especially those who did not answer all of the questions in the questionnaires. I would have never learned multiple imputation if it was not for you.

Finally, Aleksandra, thank you for your support in the past four years. You were there for me even at the times I have not deserved that.

Ana, this was all for you… Thank you for walking the Earth…

Love,

Marko Tromsø, June 2018

(4)

Summary

Background: An association between coffee consumption and the risk of cancer has long been investigated. This is particularly true for most commonly diagnosed cancer types such as, breast, endometrial, colorectal, lung, and ovarian cancer. Studies on the association between heavy coffee consumption and risk of less frequently diagnosed cancers are scarce. Coffee consumption among Scandinavian population is high, thus this is a favorable population in which to study the impact of coffee on cancer incidence.

Objectives: We aimed to quantify the association between total coffee consumption and the risk of breast, colorectal, lung, ovarian, and cancer at any site. (Paper 1) Further objective was to assess the association between filtered, boiled, and total coffee consumption and the risk of bladder, esophageal, kidney, pancreatic, and stomach cancer. (Paper 2) Finally, we aimed to summarize the existing evidence on association between coffee consumption and the risk of endometrial cancer. (Paper 3) Methods: In paper 1, baseline information on total coffee consumption was collected from 91 767 women in the Norwegian Women and Cancer Study. These information was updated from the follow- up survey conducted 6-8 years after baseline. In paper 2, we used data from the Norwegian Women and Cancer Study and the Northern Sweden Health and Disease Study. Information on coffee

consumption was available for 193 439 participants. Data on cancer incidence were obtained through linkage to the Norwegian Cancer Registry and the regional branch of the Swedish Cancer Registry.

We used multivariable Cox proportional hazards models to calculate hazard ratios (HR) with 95%

confidence intervals (CI) for the investigated cancer sites by category of total, filtered, and boiled coffee consumption. In paper 3, we searched online databases for studies published up to August 2016 that aimed to investigate the association between coffee consumption and the risk of endometrial cancer. We estimated summary relative risks (RR) for cohort studies and odds ratios (OR) for case- control studies with 95% CI by applying random effects meta-analyses. Dose-response analyses were conducted by using generalized least square trend estimation.

Results: We found and a 9% reduced risk of cancer at any site (95% CI 3%-14%, ptrend=0.03) in women who drank more than 3 and up to 7 cups/day, compared to women who drank ≤1 cups/day.

Consumption of more than 3 and up to 7 cups/day was associated with 17% reduced risk of colorectal cancer (95% CI 2%-30%) with no evidence of linear relationship (ptrend=0.10). A significantly

increased risk of lung cancer observed with a coffee consumption of >7 cups/day (HR=2.01, 95%CI 1.47-2.75, ptrend<0.001) was most likely caused by residual confounding due to smoking, as no statistically significant association was observed in never smokers (>5 cups/day HR=1.42, 95%CI

(7)

0.44-4.57, ptrend=0.30). No significant association was found between coffee consumption and the risk of breast or ovarian cancer.

Heavy filtered coffee consumption (≥4 cups/day) was associated with a reduced risk of pancreatic cancer (HR=0.74, 95% CI 0.57-0.95, ptrend=0.01). We did not observe significant associations between total and filtered coffee consumption and the risk of bladder, esophageal, kidney, and stomach cancer sites. We found some indications of a positive association between moderate boiled coffee

consumption and the bladder cancer risk in women (HR=1.58, 95% CI 1.03-2.05), but this finding was not supported by the results from corresponding dose-response analysis (ptrend=0.56). However, we found an increased risk of bladder cancer among never smokers who were heavy filtered or total coffee consumers, and an increased risk of stomach cancer in never smokers who were heavy boiled coffee consumers.

We identified twelve prospective cohort studies and eight case-control studies eligible for inclusion in the meta-analysis of coffee consumption and endometrial cancer risk. The summary RR for highest compared with lowest coffee intake was 0.73 (95% CI: 0.67–0.80, I² = 36.7%) in the combined analysis of cohort and case-control studies. The corresponding RR among cohort studies was 0.76 (95% CI: 0.69–0.83, I² = 40.5%), and the meta-OR of 0.63 (95% CI: 0.53–0.76, I² = 0%) was found after pooling the results from case-control studies. One-cup of coffee increment per day was associated with 3% (95% CI: 2% -4%) lower risk of endometrial cancer in cohort studies and 12%

(95% CI: 5% -18%) in case-control studies. After pooling the results from five cohort studies, the association remained significant only in women with body mass index over 30.

Conclusion: Increase in total coffee consumption was found to modestly reduce risk of cancer at any site, whereas increased filtered coffee consumption was associated with lower risk of pancreatic cancer. Our data suggest that increased coffee consumption does not affect the risk of breast, esophageal, colorectal, kidney, lung, and ovarian cancer sites. The positive associations between coffee consumption and the risk of bladder and stomach cancer were found in never smokers. The results from the meta-analysis indicate a protective effect of coffee consumption on the risk of endometrial cancer, particularly in women with obesity.

(8)

Sammendrag

Bakgrunn: En rekke studier har undersøkt sammenhengen mellom kaffeinntak og risiko for kreft, spesielt for de hyppigst forekommende kreftformene som bryst-, livmor-, tarm-, lunge- og

eggstokkreft. Derimot fins det få studier av kaffeinntak og risiko for sjeldnere kreftformer.

Kaffeinntaket i den skandinaviske befolkningen er høyt, og dette er et fortrinn når man ønsker å studere betydningen av kaffeinntak for kreftrisiko.

Formål: Målet var å estimere sammenhengen mellom totalt kaffeinntak og risiko for kreft i bryst, tarm, lunge og eggstokk, samt total kreft. (Artikkel 1) Videre ønsket vi å studere betydningen av tilberedningsmetode (filterkaffe og kokekaffe i tillegg til totalt kaffeinntak) for kreft i blære, spiserør, nyre, bukspyttkjertel og mage. (Artikkel 2) Til slutt hadde vi som formål å oppsummere eksisterende studier av kaffeinntak og risiko for livmorkreft. (Artikkel 3)

Metoder: I artikkel 1 ble informasjon om totalt kaffeinntak innhentet fra 91 767 kvinner i den norske studien Kvinner og kreft. Denne informasjonen ble oppdatert via et nytt spørreskjema 6-8 år senere. I artikkel 2 brukte vi data fra Kvinner og kreft og den nord-svenske helse- og sykdomsstudien NSHDS, hvor informasjon om kaffeinntak var tilgjengelig for 193 439 deltakere. Informasjon om kreftinsidens ble skaffet ved hjelp av koblinger til Kreftregisteret i Norge og det regionale kreftregisteret i Sverige.

Vi brukte multivariable Cox’ regresjonsmodeller for proporsjonale hasarder for å beregne hasardratio (HR) med 95% konfidensintervall (KI) for de inkluderte kreftformene ved inntak av total-, filter- og kokekaffe. I artikkel 3 søkte vi i nettbaserte databaser etter studier publisert frem til august 2016, hvor formålet var å undersøke sammenhengen mellom kaffekonsum og risiko for livmorkreft. Vi estimerte samlet relativ risiko (RR) for kohortstudier og oddsratio (OR) for kasus-kontrollstudier med 95% KI ved å anvende randomiserte meta-analyser. Dose-responseffekter ble estimert ved en «generaliserte minste kvadrat» trendanalyse.

Resultater: Vi fant en 9% redusert risiko for kreft totalt (95% KI 3% -14%, p-verdi for trend = 0,03) hos kvinner som drakk mer enn 3 og opptil 7 kopper per dag, sammenlignet med kvinner som drakk 1 kopp eller mindre per dag. Inntak av mer enn 3 og opptil 7 kopper per dag var assosiert med 17%

redusert risiko for tarmkreft (95% KI 2% -30%) uten noen lineær trend (p = 0,10). Vi observerte en signifikant økt risiko for lungekreft ved kaffeinntak på mer enn 7 kopper per dag (HR = 2,01, 95% KI 1,47-2,75, p for trend <0,001), som mest sannsynlig var forårsaket av restkonfundering fra røyking, da ingen statistisk signifikant sammenheng ble observert hos aldri-røykere (> 5 kopper / dag HR = 1,42, 95% KI 0,44-4,57, p for trend = 0,30). Ingen signifikant assosiasjon ble funnet mellom kaffeinntak og risiko for bryst- eller eggstokkreft.

(9)

Høyt inntak av filterkaffe (≥ 4 kopper / dag) var assosiert med redusert risiko for kreft i

bukspyttkjertelen (HR = 0,74, 95% KI 0,57-0,95, p for trend = 0,01). Vi observerte ingen signifikante sammenhenger mellom inntak av total-, filter- eller kokekaffe og risikoen for kreft i blære, spiserør, nyre eller mage. Vi fant indikasjoner på en positiv sammenheng mellom moderat inntak av kokekaffe og blærekreft hos kvinner (HR=1.58, 95% KI 1.03-2.05), men dette funnet støttes ikke av resultatene fra dose-responsanalysen. (p for trend = 0.56). Imidlertid fant vi en økt risiko for blærekreft blant røykere med høyt inntak av filterkaffe eller kaffe totalt, og en økt risiko for magekreft hos aldri- røykere med høyt inntak av kokekaffe.

Vi identifiserte tolv prospektive kohortstudier og åtte kasus-kontrollstudier som var kvalifisert for inkludering i meta-analysen av kaffeinntak og risiko for livmorkreft. Samlet RR for høyeste sammenlignet med laveste kaffeinntak var 0,73 (95% KI: 0,67-0,80, I2 = 36,7%) i den kombinerte analysen av kohort- og kasus-kontrollstudier. Tilsvarende RR blant kohortstudier var 0,76 (95% KI:

0,69-0,83, I2 = 40,5%), og meta-OR 0,63 (95% KI: 0,53-0,76, I2 = 0%) ble funnet etter å ha samlet resultatene fra kasus-kontrollstudier. En kopp økning i kaffeinntak per dag var assosiert med 3% (95%

KI: 2% -4%) lavere risiko for livmorkreft i kohortstudier og 12% (95% KI: 5% -18%) i kasus- kontrollstudier. Etter å ha samlet resultatene fra fem kohortstudier, forble sammenhengen signifikant bare hos kvinner med kroppsmasseindeks over 30.

Konklusjoner: Vi fant en moderat reduksjon i risiko for kreft totalt ved høyt inntak av kaffe, mens økt inntak av filterkaffe var forbundet med lavere risiko for kreft i bukspyttkjertelen. Våre data tyder på at økt kaffeinntak ikke påvirker risikoen for kreft i bryst, spiserør, tarm, nyre, lunge eller

eggstokker. En positiv assosiasjon mellom kaffeinntak og risiko for blære- og magekreft ble funnet bare hos røykere. Resultatene fra meta-analysen indikerer en beskyttende effekt av kaffeinntak på risikoen for livmorkreft hos kvinner med fedme.

(10)

List of papers

This thesis is based on the following papers, hereafter referred in the text as Papers 1,2, and 3.

PAPER 1

Lukic M, Licaj I, Lund E, Skeie G, Weiderpass E, Braaten T. Coffee consumption and the risk of cancer in the Norwegian Women and Cancer (NOWAC) Study. Eur J Epidemiol. 2016;31(9):905- 16.

PAPER 2

Lukic M, Nilsson LM, Skeie G, Lindahl B, Braaten T. Coffee consumption and risk of rare cancers in Scandinavian countries. Eur J Epidemiol. 2018;33(3):287-302.

PAPER 3

Lukic M, Guha N, Licaj I, van den Brandt PA, Stayner LT, Tavani A, et al. Coffee Drinking and the Risk of Endometrial Cancer: An Updated Meta-Analysis of Observational Studies. Nutr Cancer.

2018;70(4):513-28.

(11)

Abbreviations

CYP Cytochrome P450

NAT2 N- acetyltransferase 2

IARC International Agency for Research on Cancer

OR Odds Ratio

CI Confidence Interval

WCRF World Cancer Research Fund

EPIC European Prospective Investigation into Cancer and Nutrition

NOWAC Norwegian Women and Cancer Study

NSHDS Northern Sweden Health and Disease Study VIP Västerbotten Intervention Programme

MONICA Monitoring Trends and Determinants in Cardiovascular Disease NSDD Northern Sweden Diet Database

BMI Body mass index

HR Hazard ratio

RR Risk ratio

FFQ Food-frequency questionnaires

MCAR Missing completely at random

MAR Missing at random

MNAR Missing not at random

MICE Multiple imputation by chained equations

CPU Central processing unit

PH Proportional hazards

DNA Deoxyribonucleic acid

(12)

1 Introduction

1.1 Coffee

1.1.1 Biological classification and history

The genus of plants known as Coffea consists of over 70 species (1, 2). Most of the Coffea species can be found in the Equatorial region with the optimum altitude for growth being between 1300 meters and 1600 meters (3, 4). Our morning coffee is most likely made either of Coffea arabica or Coffea canephora (s. Coffea robusta), the two species dominating the coffee industry as they account for approximately 60% and 40% of the world coffee consumption, respectively. Other less grown species include Coffea liberica and Coffea excelsa, which contribute to up to 2% of world’s gross production and are considered to be of poorer overall quality (1, 2, 5).

It is believed that coffee consumption originates either from Ethiopia or Yemen (6). The modern way of processing coffee beans by roasting and grounding was reported first in the Yemeni city Mocca in early 15^th century (7). Even though the Ottoman Turks who had ruled Yemen tried to confine coffee use strictly within its borders, coffee consumption has spread via Istanbul to Venice in the beginning of the 17^thcentury. Some decades later, coffee appeared first in France, and later in Vienna and England, reaching North America presumably in 1668 (6, 7).

The first mention of coffee in Norway was recorded in 1694. During the early 18^th century, coffee supressed alcoholic beverages as a “social lubricant” among the upper class of Norwegian society (8).

By the late 18^th century, approximately 250-300 gr of unroasted coffee per person was imported to Norway yearly through Copenhagen (8). The amount increased ten-fold by the mid-19^th century when, Friele, a merchant from Bergen, started trading salted cod with coffee producers in Brazil, importing over 900 tonnes of coffee per year (9). At the time, coffee was made in copper kettles by boiling, with fish-skin and/or hot charcoal being commonly added into kettles before serving (10). Due to an increased coffee consumption among Norwegians throughout 1850s and 1860s, health authorities rose their concerns about its health effects for the first time (11).

By the end of the 19^th century, yearly coffee consumption in Norway had reached almost 4 kg per capita. This figure has fluctuated during the 20^th century, rising up to 6.5 kg per person during the prohibition between 1916 and 1927, and plummeting to only 50 grams per week during the German occupation of Norway (12). After the World War 2, coffee consumption in Norway increased again, with 50% of coffee nowadays being imported from Brazil, and 25% from Colombia and Guatemala (12). Between 1997 and 2011, average consumption of coffee in Norway reached 9.4 kg/year per capita, and only Finland among Nordic countries was ahead with 11.7 kg (13). In the past few

(13)

decades, the preferred brewing method has changed among Norwegians. Filtered coffee is now considered by far the most popular brewing method surpassing both boiled and instant coffee as the alternatives (14). As from the late 1990’s and throughout 2000’s other types of coffee drinks such as macchiato, espresso, cappuccino, or café latte has become increasingly popular, while the

consumption of decaffeinated coffee in Norway remains uncommon to this day (15).

In Sweden, the average roasted coffee consumption between 2006 and 2016 was stable and averaged 7.95 kilograms per capita, which is a slight decrease compared to the period between 1997 and 2005 (8.2 kilogram per capita) (13, 16). The consumption peaked at 2010 reaching 8.8 kilogram per person. However, the following year, the consumption of only 7.3 kilogram of roasted coffee was reported per capita. The data from 2016 have shown the consumption of roasted coffee in Sweden amounted to 8.1 kilograms per capita. The total annual volume of roasted coffee consumed in Sweden averages 80 thousand metric tons (16).

1.1.2 Coffee constituents

Coffee as a beverage consists of numerous components which concentration and bioavailability depend on coffee type, roasting and brewing methods (17).

Caffeine

Caffeine is an alkaloid also known as 1,3,7-trimethylxanthine,with a chemical formula C8H10N4O2

(Figure 1) (18). It is considered the most frequently consumed behaviorally active substance in the world (17-19), and other than in coffee, it can be found in tea, cocoa beverages, chocolate, energy drinks, and pre-workout beverages. On average, Coffea robusta contains more caffeine compared to Coffea arabica (19-21 mg/g vs. 10-12 mg/g, respectively) (20, 21). The content of caffeine seems to be dependent on brewing method, with boiled coffee having higher caffeine concentration per 100 ml compared to filtered coffee (22).

Figure 1: Chemical structure of caffeine. Reprinted from: National Center for Biotechnology

(14)

The peak plasma concentration of caffeine is reached 15-20 minutes after oral ingestion. It is absorbed completely by the stomach and small intestine, and it is able to pass all biological membranes (24, 25).

The primary metabolism of caffeine in liver involves CYP1A2 (Cytochrome P450 1A2) enzyme that belongs to a cytochrome P450 oxidase enzyme system, and by N-acetyltransferase 2 enzyme (NAT2) (25-30). Caffeine has a half-life of approximately 5 hours in healthy adults and is mainly metabolized to paraxanthine (70-80%), and only a small fraction is excreted unchanged in the urine (25, 31).

Caffeine exhibits the ability to improve physical (25, 32) and cognitive performance, such as vigilance and decreased reaction times (25). It is also shown to manage the symptoms of Parkinson’s disease (33, 34) and to have antioxidant properties (35, 36). Adverse effects of caffeine, such as increase of anxiety or jitteriness, are mainly the result of excessive amount of caffeine consumed and can also be a consequence of caffeine withdrawal (37, 38).

Diterpenes

Two diterpenes found in a coffee bean are cafestol and kahweol which were found to have anticarcinogenic properties (39, 40) (Figure 2). Studies have shown that kahweol could induce apoptosis in human leukemia cells (41), to reduce gentoxicity in hepatoma cells (42), and to induce synthesis of endogenous antioxidants (43), whereas both cafestol and kahweol were reported to induce apoptosis in human malignant pleural mesothelioma (44). On the other hand, coffee diterpenes were also found to increase serum cholesterol and might cause extracellular accumulation of low-density lipoproteins (45-48). Filtered coffee, as one of the methods of preparing coffee beverage, was found to have lower levels of diterpenes compared to boiled, espresso, or French press coffee (49-51), as the paper filter manages to block the passage of fine particles into the brew, and diterpenes being retained by the spent coffee (52, 53).

Figure 2: Chemical structure of cafestol and kahweol. Reprinted from: National Center for

(15)

Chlorogenic acids

Chlorogenic acids are polyphenolic compounds, with 5-caffeoylquinic acid being its major component found in coffee, and neochlorogenic, cryptochlorogenic^,feruloylquinic, and dicaffeoylquinic acids also present in large quantities (18, 56-58) (Figure 3). It is one of the coffee ingredients with very high antioxidant activity (57, 59-63). In addition, it has been speculated that chlorogenic acids could decrease the risk of some cancers, and the proposed mechanism is that it reduces glucose levels in the blood and increases insulin sensitivity (64-67). Concentration of chlorogenic acids in a regular cup of coffee depends on type of coffee bean, roasting temperature, and brewing method (22, 57, 58, 68-70).

Caffea robusta was found to have a higher content of chlorogenic acid compared to Caffea arabica, and therefore the higher antioxidant activity (22, 57, 61). Similarly, more pronounced antioxidant activity of boiled compared to filtered coffee could be the result of higher concentrations per cup of not only diterpenes but chlorogenic acids as well (22).

Coffee Maillard Reaction Products

During the process of high temperature roasting of green coffee beans, high molecular weight, polymeric chemicals named melanoidins are created as products of Maillard reaction (21, 72). The structures of these chemicals responsible for the taste and aroma of coffee beverage are difficult to determine, as melanoidins are modified to less known structures from polysaccharides, proteins, and phenolics (21, 73, 74). The products of Maillard reactions were also shown to have antioxidant properties (72, 75, 76), and the mechanism of antioxidant activity includes scavenging hydroxyl and proxy radicals, and breaking the radical chain (77). In vivo, these molecules were found to have a protective effect against oxidative stress in human hepatoma cells (78), to increase survival of human neuroblastoma cells from oxidative damage (75), and to suppress liver oxidative stress in rats (79).

(16)

1.1.3 Cancer incidence in Norway

The incidence rate for all cancer sites combined has increased by 4.4% among Norwegian women in the five-year period from 2012 to 2016 compared to the period between 2007 and 2011. Breast, colorectal, and lung cancer remain the three most frequently diagnosed cancers in Norway and worldwide (80, 81). In Norway, age-standardized incident rate of breast cancer per 100 000 person- years has increased by 7% from 2007–11 to 2012–16. Comparing the same periods, the incidence rate of lung cancer has increased by 9.4%. Women in Norway have one of the highest incidences of colorectal cancer in the Nordic countries, with 1588 new cases of colon and 517 new cases of rectal cancer diagnosed in 2016. The incidence of colon cancer has increased almost three-fold between 1955 and 2014, while rectal cancer rates have stabilized after 1990s. On the other hand, a reduction in rates was recorded for ovarian (-7.4% between 2007–11 and 2012–16) and endometrial cancer (- 5.9%). In 2016, age-standardized (Norwegian standard) incident rate of esophageal cancer was 2.4 per 100 000 women compared to 8.5 per 100 000 men. Similarly, higher rates are observed in Norwegian men compared to women for stomach (12.1 per 100 000 vs. 5.0 per 100 000), pancreatic (15.3 per 100 000 vs. 11.5 per 100 000), kidney (22.9 per 100 000 vs. 10.1 per 100 000), and urinary tract (52.8 per 100 000 vs. 16.9 per 100 000) cancer sites (80).

1.1.4 Cancer incidence in Sweden

From 2007 to 2016, the breast cancer incidence rates in Sweden have increased from 144 per 100 000 to 163.4 per 100 000 women, and breast cancer remained the most frequently diagnosed cancer among Swedish women. In Swedish men, prostate cancer is the most commonly diagnosed cancer with about 10,500 new cases being diagnosed in 2016. In the same year, the incidence rate of lung cancer was similar in men (35.5 per 100 000) and women (35.9 per 100 000) whereas higher rates of colon cancer were reported in men (47.1 per 100 000 vs. 42.1 per 100 000). Cancer of upper digestive tract was also more common in men (19.8 per 100 000) compared to women (9.6 per 100 000), which was also true for cancer of liver, pancreas and hepatobiliary tract (25.9 per 100 000 vs. 19.6 per 100 000), and cancer of kidney and urinary tract (62.2 per 100 000 vs. 23.5 per 100 000) (82).

1.2 Coffee consumption and the risk of cancer – the story so far

Because of the vast popularity of coffee beverages worldwide, any causal association between coffee consumption and chronic diseases would have a significant public health impact. The amount of evidence from epidemiological/health effect studies which had the aim to explore the association between coffee consumption and the risk of cancer, was mounting in the past two decades. In 2016,

(17)

the International Agency for Research on Cancer (IARC) published the Monograph in which they have evaluated the carcinogenicity of drinking coffee, maté, and very hot beverages, based on the available literature. In a summary of the final evaluations, the authors of the monograph stated that coffee drinking was not classifiable according to its carcinogenicity in humans (Group 3): “there was inadequate evidence of carcinogenicity in experimental animals and evidence suggesting lack of carcinogenicity in humans for cancers of the pancreas, female breast, and prostate, liver, and uterine endometrium, with inverse associations noted for the latter two sites” (83).

Coffee consumption and bladder cancer

The two epidemiological studies from Norway that aimed to explore the association between coffee drinking and the risk of bladder cancer included 16 555 and 43 000 Norwegian men and women, respectively (84, 85). The authors of both studies did not observe increase in the bladder cancer incidence among heavy coffee consumers. The recent meta-analysis of 25 case-control and five cohort studies revealed an overall meta-odds ratio (OR) of 1.33 (95% confidence interval (CI) 1.19-1.48) between coffee consumption and the bladder cancer risk (86). Finally, in the 2016 Monograph, the authors concluded there was no consistent evidence of an association between coffee consumption and bladder cancer risk, and that positive associations observed in some studies are most likely due to inadequate adjustment for smoking exposure, a known risk factor (83).

Coffee consumption and breast cancer

The two most recent meta-analyses on coffee consumption and the risk of breast cancer were published in 2013 and included 37 (87) and 26 (88) observational studies. Jiang et al. reported an inverse association for postmenopausal women and women that were BRCA1 mutation carriers (87), whereas Li et al. concluded that coffee drinking was not associated with the breast cancer risk (88). In Norway, Stensvold and Jacobsen found a statistically non-significant positive association between coffee consumption and the risk of most frequently diagnosed cancer in Norway (85). As mentioned above, the authors of IARC’s Monograph concluded that evidence from the available studies indicate lack of carcinogenicity for the female breast (83). The World Cancer Research Fund (WCRF) in their Continuous Update Project on how diet, nutrition, physical activity and body weight affect cancer risk and survival, concluded that the evidence on coffee and breast cancer risk are limited (89).

Coffee consumption and colorectal cancer

In the recent meta-analysis, Gun et al. found a 7% decreased risk of colon cancer for an increase in coffee intake by 4 cups per day (95% CI 1%-12%). In conclusion, the authors have stated that coffee consumption of at least 5 cups of coffee per day was inversely associated with the risk of colorectal

(18)

of protective effect of coffee drinking on the risk of colorectal adenoma (83), whereas the authors of WCRF’s Continuous Update Project conclude that the evidence are limited/inconclusive (91). Among Norwegian women, the reported inverse association was not statistically significant for either colon or rectal cancer sites (85).

Coffee consumption and endometrial cancer

In two most recently published meta-analyses, coffee consumption was found to be inversely

associated with the risk of endometrial cancer, and that the observed effect may be modified by body mass index (BMI), previous use of hormone replacement therapy (92), and menopausal status (93).

The overall conclusion from these meta-analyses are in line with the conclusion from IARC’s Monograph (83), and from the Continuous Update Project (94).

Coffee consumption and esophageal cancer

There are little evidence indicating that coffee is associated with esophageal cancer (95, 96). However, there are strong evidence that consumption of hot beverages increases the risk of esophageal cancer (83, 96), based on the findings from one meta-analysis (95) and two studies from South America (97, 98).

Coffee consumption and lung cancer

The results from a 2016 meta-analysis of 17 observational studies suggested that coffee consumption is associated with an increased risk of lung cancer (99), which was previously found in the study by Stensvold and Jacobsen (85). Galarraga et al. speculated that this association would be non-existent if effect of smoking is properly accounted for in the analyses, and indicated the importance of pooled analyses with large study groups of never smokers (100). In the WCRF’s Continuous Update Project report, the authors concluded that the evidence on coffee and lung cancer incidence are

limited/inconclusive (101).

Coffee consumption and ovarian cancer

No association between coffee intake and incident ovarian cancer is was found in the study from the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort, and from a meta-analysis by the same authors (102). The same conclusion was made in the umbrella review of meta-analyses of coffee consumption and multiple health outcomes (103), and from the results of the Continuous Update Project (104).

(19)

Coffee consumption and pancreatic cancer

A 2012 meta-analysis revealed no apparent association between high coffee intake and the risk of pancreatic cancer (105). However, in more recent meta-analysis of 20 cohort studies, Ran et al.

concluded that high coffee consumption was in fact associated with the decreased pancreatic cancer risk (106). This finding was not supported by the IARC’s 2016 Monograph, nor by the authors of the Continuous Update Project (83, 107). Similarly, no significant relationship was found in two studies from Norway (84, 85).

Coffee consumption and renal cancer

In the Norwegian study from 1986, Jacobsen et al. found a strong inverse association between coffee intake and the renal cancer risk (84). This finding was later refuted by Stensvold and Jacobsen, who found no such relationship based on the data from 43 000 Norwegian men and women (85). This later finding was also backed up by the conclusion from the Continuous Update Project (108).

Coffee consumption and stomach cancer

Five meta-analyses published in 2015 and 2016 reported conflicting results regarding the relationship between coffee intake and the incident stomach cancer, ranging from decreased risk (109), no

association for stomach cancer risk overall (110) and in the population outside of United States (111, 112), to an increased risk (113). Limited evidence of an association was reported by the Continuous Update Project (114), which was also found previously in the two studies from Norway (84, 85).

(20)

2 Aims

The overall aim of the thesis is to explore the association between coffee consumption and the risk of cancer.

PAPER 1

The aim was to investigate the relationship between heavy coffee consumption and the risk of breast, colorectal, ovarian, and lung cancers, as well as cancer at any site (overall cancer risk), in the

Norwegian Women and Cancer (NOWAC) study.

PAPER 2

By pooling the data from two Nordic cohorts: the Norwegian Women and Cancer (NOWAC) Study and the Northern Sweden Health and Disease Study (NSHDS), we explored the association between filtered, boiled, and total coffee consumption and the risk of bladder, esophageal, kidney, pancreatic, and stomach cancer.

PAPER 3

We aimed to summarize the available evidence on the association between coffee intake and endometrial cancer risk by conducting an updated dose-response meta-analysis of observational studies.

.

(21)

3 Material and methods

3.1 Papers 1 and 2

3.1.1 The Norwegian Women and Cancer Study

The NOWAC study is population based cohort study that involves approximately 172 000 women from all over Norway. It was initiated in 1991 with the primary aim to investigate risk factors for breast cancer. The NOWAC Study has been described in detail by Lund et al (115). In short, random samples of Norwegian women aged 30-70 years was drawn from the Norwegian Central Population Registry and were invited to participate. The enrolment of women was conducted in three waves (1991, 1995-1997, and 2003-2007) (Figure 4). The food frequency questionnaire (FFQ) was added during the second wave of enrolment, whereas the collection of blood samples was initiated in 2003.

More than 172 000 accepted and completed a questionnaire regarding their lifestyle, diet, and health status (overall response rate: 52.7%), and the NOWAC biobank consists of over 60 000 blood samples. All the participants received two follow-up questionnaires within 5 to 8 years between each follow-up. The NOWAC Study was approved by the Regional Committee for Medical Research Ethics and the Norwegian Data Inspectorate. All women gave written informed consent.

3.1.2 Northern Sweden Health and Disease Study

The NSHDS contains survey and biobank data from about 166 000 men and women in northernmost Sweden. It was initiated in 1985 and includes three population-based sub-cohorts: the Västerbotten Intervention Programme (VIP) cohort (participants aged 30, 40, 50 or 60 years), the MONICA (Monitoring Trends and Determinants in Cardiovascular Disease) cohort (participants aged 25-74 years), and the mammography screening cohort (women aged 50-69 years). Dietary data administered by the Northern Sweden Diet Database (NSDD) are available in the VIP and MONICA cohorts. The participants of the ongoing VIP cohort that runs in the Northern Sweden’s Västerbotten county have undergone extensive health examination and have also answered questionnaires regarding diet, lifestyle and health conditions. The mean participation rate has been around 60% (116).

The MONICA cohort includes inhabitants in the Västerbotten and Norrbotten counties randomly selected from the population registries that were updated every fourth to fifth year (117). The participation rate over the years of recruitment varied from 62% to 81%. The questionnaires used in the MONICA cohort are very similar to those used in the VIP cohort (116). The coffee study within the NSHDS was approved by the Regional Ethical Review Board of Northern Sweden, and all participants gave written informed consent.

(22)

Figure 4: Cohort enrollment in the NOWAC study

(23)

3.1.3 Study population

For paper 1, we used the information from the questionnaires of women enrolled in 1991-1992, 1996- 1997, 2003, and 2004. These women completed baseline food frequency questionnaires in 1998, 1996- 1997, 2003, and 2004, respectively. The information collected during the first wave (1991-1992) were not used in the study, as the version of questionnaires that was sent out did not include questions regarding diet. However, as the women enrolled during the first wave of data collection answered dietary questions in the follow-up survey in 1998, we used this information as baseline data for the present study. In total, 98 405 women answered the question regarding their coffee consumption.

Women with prevalent cancer other than non-melanoma skin cancer at baseline and those who emigrated or died before the start of follow-up (N=4395), those who were diagnosed with cancer after they emigrated (N=9), and those with total energy intake above 15 000 kJ or below 2500 kJ per day (N=619), and those with missing information on coffee intake at baseline (N=1615) were excluded from the study. Our final analytical study sample consisted of 91 767 women. Follow-up information were available from 79 461 of these women, who received the follow-up questionnaire before the end of the study, 6-8 years after baseline data collection. The rest of the women (N=12 306) received the baseline questionnaire in 2004, while the follow-up questionnaire was sent out to them after the study had ended.

In paper 2, the NOWAC cohort initially included 101 320 women, while the NSHDS cohort

contributed 103 256 participants. As in paper 1, the data collected during 1991-1992 were not used in the study. After exclusion of those participants with prevalent cancer other than non-melanoma skin cancer at baseline, those who emigrated or died before the start of follow-up (N=8101), those that had missing information on coffee consumption (N=1615), and those with total energy intake above 15 000 kJ or below 2500 kJ/day (N=1362), the final sample consisted of 145 247 women and 48 192 men.

3.1.4 Information on coffee consumption and covariates

In the NOWAC cohort, the same question regarding coffee consumption was asked at baseline and at follow-up: “How many cups of each kind of coffee (boiled, filtered, instant) did you usually drink during the past year?” Women could choose from the following answers: never/seldom, 1-6 cups/week, 1 cup/day, 2-3 cups/day, 4-5 cups/day, 6-7 cups/day, and ≥8 cups/day for each brewing method. For paper 1, we calculated total coffee consumption by summing the frequencies of each of the brewing methods, and categorized it as ≤1 cup/day (light consumers), more than 1 up to 3 cups/day (low moderate consumers), more than 3 up to 7 cups/day (high moderate consumers), and >7 cups/day (heavy consumers).

(24)

In the NSHDS cohort, the participants were asked the number of occasions on which they consumed filtered or boiled coffee with the following alternatives: never, a few times/year, 1-3 times/month, 1 time/week, 2-3 times/week, 4-6 times/week, 1 time/day, 2-3 times/day, and >4 times/day.

For the purpose of the paper 2, we assumed that 1 time/day from the NSHDS cohort matched 1 cup/day from the NOWAC cohort. Total coffee consumption by summing the frequencies of filtered, boiled, and instant coffee in the NOWAC cohort and filtered and boiled coffee in the NSHDS cohort.

Filtered, boiled, and total coffee consumption was then categorized as light consumption (≤1 cup/day), moderate consumption (>1-< 4 cups/day), and heavy consumption (≥4 cups/day).

In paper 1, the following information were collected at both baseline and follow-up: BMI (body mass index), physical activity, alcohol consumption, total energy intake, and use of hormone replacement therapy (never, former, current), smoking status (never, former, current), and number of pack-years (calculated as number of cigarettes smoked/day divided by 20 and multiplied by years of smoking).

We categorized as former smokers at follow-up those women who reported they were current or former smokers at baseline and non-smokers at follow-up (N=1608).

In paper 2, we also utilized information on smoking status (never, former, current), alcohol consumption, BMI, total energy intake, and self-reported history of diabetes.

3.1.5 Cancer incidence, death, and emigration

We used the unique 11-digit personal number assigned to every legal resident in Norway, and the unique 12-digit personal number assigned to residents in Sweden to obtain information on cancer incidence, death, and emigration through linkage to the Norwegian Cancer Registry, the Cause of Death Registry, the Norwegian Central Population Register, the regional branch of the Swedish Cancer Registry, and Swedish National Cause-of-death Registry.

The 7th Revision of the International Statistical Classification of Diseases, Injuries and Causes of Death was used to classify breast (170.0-170.9), colorectal (153.0-154.0), ovarian (175.0-175.9), and lung (162.0-162.1) (paper 1), and bladder (181.0-181.9), esophageal (150.0-150.9), kidney (180.0- 180.9), pancreatic (157.0-157.9), and stomach (151.0-151.9) cancer cases (paper 2) in the national and regional cancer registries.

3.1.6 Statistical analysis

In paper 1, we applied baseline information on coffee consumption and covariates until potential follow-up information, date of diagnosis of any incident cancer other than non-melanoma skin cancer, death, or emigration, whichever occurred first. Thereafter, we updated information on coffee

(25)

consumption and smoking variables until diagnosis of any incident cancer other than non-melanoma skin cancer, until death, emigration or the end of the study period (31 December 2013), whichever occurred first. In paper 2, person-years were calculated from start of follow-up until diagnosis of any incident cancer other than non-melanoma skin cancer, death, emigration, or the end of the study period (31 December 2014), whichever occurred first.

Cox proportional hazards regression models were used to calculate hazard ratios (HRs) for developing breast, colorectal, ovarian, or lung cancer, as well as cancer at any site other than non-melanoma skin cancer in paper 1, and bladder, esophageal, kidney, pancreatic, or stomach cancer in paper 2, with 95% CIs for each coffee consumption group. Light consumers (i.e., those drinking ≤1 cup/day), were used as the reference group, and attained age was used as the underlying time scale in both papers. In paper 1, all models were stratified by questionnaire subcohorts, (i.e. during which of the three waves of data collection a participant was included in the cohort), and by study cohort (NOWAC/NSHDS) in paper 2.

In paper 1, analyses for each cancer site were adjusted for known risk factors for specific cancer type in the preliminary, complete-case analysis: menopausal status (premenopausal/postmenopausal), smoking status (never, former, current), age at smoking initiation (<20, ≥20 years), number of pack-years (≤14, 15-19, ≥20), exposure to cigarette smoke during childhood (yes/no),

duration of education (≤9, 10-12, 13-16, ≥17 years), body mass index (BMI, ≤18.49, 18.5- 24.9, 25-29.9, and ≥30 kg/m²), physical activity level (1-4, 5-6, 7-10), alcohol consumption (0, 0.1-3.99, 4-9.99, ≥10 g/day), number of children (0, 1-2, ≥3), age at first birth (<20, 20-24, 25-29, ≥30 years), ever use of oral contraceptives (yes/no), duration of oral contraceptive use in years (continuous), use of hormone replacement therapy (never, former, current), maternal history of breast cancer (yes/no), total energy intake (tertiles, kJ/day), intake of fibers (<=20,

>20 g/day), intake of processed meat (continuous, g/day), intake of red meat (<=10, 10.01-20,

>20, g/day), height (continuous, cm), and participation in mammography screening (yes/no).

For the analyses of colorectal, lung, and cancer at any site, smoking exposure was modelled as five- categories variables, which included the information on smoking status, age at smoking initiation, and number of pack-years. We built the final models by assessing if the removal of the covariate lead to a change in the regression coefficients of at least 10% in any of the coffee consumption groups. In paper 2, we adjusted for a variety of a priori selected risk factors: sex, smoking status (never, former, current), BMI (≤18.49, 18.5-24.9, 25-29.9, and ≥30 kg/m²), alcohol consumption (0, 0.1-3.99, 4- 9.99, ≥10 g/day), and self-reported history of diabetes (yes/no). In addition, the analyses of filtered coffee were adjusted for boiled coffee consumption and vice-versa.

(26)

The proportional hazards assumption was assessed by testing an interaction between coffee consumption and the logarithmic transformation of participants’ age in paper 1, and by assessing Schoenfeld residuals in paper 2. In both papers, we tested for linear trend across coffee consumption categories by assigning a median value to each category of the ordinal coffee consumption variable, which was then modeled as a continuous variable in the analysis.

We checked if smoking status, BMI, and physical activity level modified the effect of total coffee consumption on studied outcomes in paper 1, and if BMI was effect modifier in the analyses of each brewing type in paper 2. In paper 1, by using information on women that were never smokers during the entire study period, we conducted the analyses of lung cancer to deal with residual confounding due to smoking. In paper 2, we conducted subgroup analyses according to smoking status (never/ever) for all of the outcomes.

In paper 2, to assess a possible non-linear relationship between coffee consumption and the study outcomes, we modeled restricted cubic splines with four knots, with its locations based on Harrell’s recommended percentiles of the total and filtered coffee consumption (118), and positioned at the 25^th, 60^th, and 95^th percentiles of the boiled coffee distribution. We used a Wald-type test to assess if the coefficients of the second and third spline in filtered and total coffee analyses, and if the second spline’s coefficient in boiled coffee analyses were equal to zero.

The use of repeated measurement in paper 1 on coffee consumption allowed us to conduct an extensive lag analysis in order to check for possible reverse causality. First, we have performed the analyses for every outcome after excluding cancers at the corresponding sites diagnosed during the first two years of follow-up. Second, we had excluded cancer cases diagnosed during the first year of follow-up while also censoring at the time of answering the second questionnaire those cancer cases diagnosed during the first year after they received the second questionnaire. In paper 2, a possible reverse causality was assessed by excluding cancer cases of interest diagnosed during the first year of follow-up.

3.1.7 Multiple imputation

In paper 1, assuming that data was missing at random, we performed multiple imputation by chained equations to deal with missing data. Twenty imputed datasets were created with ten iterations per dataset, in order to reduce sampling variability from the imputation simulation (119). The missing values were then replaced by imputed values based on the observed information. We created imputation models for each outcome, including all of the variables from the final analysis of the specific cancer sites, and the set of variables that we assumed could increase the predictive power of the imputation procedure, regardless of whether the variable(s) were used in the multivariable Cox regression model. Any significant interaction term between coffee intake and smoking status, BMI or

(27)

physical activity level in addition to the outcome indicator, and the Nelson-Aalen cumulative hazard estimator were included as predictors in all the imputation models (120). We used predictive mean matching using the 100 closest individual observations (nearest neighbors) from which imputed values were drawn as well as logistic regression, ordinal logistic regression, and multinominal logistic

regression to impute continuous, binary, ordinal, and nominal variables, respectively.

The estimates from the twenty imputed datasets were combined using Rubin’s rules in order to obtain HRs and corresponding 95% CIs (121, 122).

In our study, the single population parameter of interest was a log HR from a Cox regression model, which will be denoted as Q. Then the multiple imputation point estimate of Q is the average of the m complete-data estimates (123):

Further, we estimated a within-imputation variance component as the average of the complete-data variance estimates (123):

Between-imputation variance is calculated by a combination of the complete-data point estimates (123):

Finally, the total variance is calculated as:

(28)

3.2 Paper 3

3.2.1 Search strategy, study selection, and data extraction

We performed a search of the electronic databases PubMed and Embase for studies published until August 2016 with the aim to quantify association between coffee intake and the risk of endometrial cancer. The following search strategy has been applied: ("coffee"[MeSH Terms] OR coffee[Text Word]) AND ("Endometrium"[Mesh] OR endometrial OR endometrium OR uterus OR corpus uteri) AND "Neoplasms"[Mesh] OR neoplasms OR cancer OR carcinogenic OR tumor) AND ("Epidemiology"[Mesh] OR "Epidemiologic Studies"[Mesh] OR "Cohort studies"[Mesh] OR "Case- control studies"[Mesh] OR case-referent OR case-control OR cohort). We included case-control, retrospective or prospective cohort studies that reported effect estimates with corresponding 95% CI for the association. We excluded the studies that did not adjust for body mass index and smoking as these are considered as key confounders.

The data were independently extracted by two authors, and any disagreements were resolved by mutual discussion. The following information were extracted from each included study: first author’s last name, publication year, country of origin, study design, age of participants at study initiation, number of participants, person-years, and endometrial cancer cases, mean, median or duration of follow-up (cohort studies), number and type of cases and controls, mean age of cases and controls (case-control studies), covariates adjusted for in the analyses, information on principal strengths and limitations, risk ratios (RRs) from cohort studies and ORs from case–control studies with 95% CIs for each coffee consumption category.

3.2.2 Statistical analysis

ORs from case-control study and RRs from cohort studies were log transformed, after which standard errors of log-transformed effect estimates were calculated to compute summary RR or OR with 95%

CI for the highest vs lowest (reference) category of coffee intake using random-effect models. The weight of each study was inversely proportional to the within-study sampling variances (124). We also conducted alternative analyses in which we first combined category specific estimates among coffee drinkers within each study into a single estimate for moderate to heavy coffee drinkers’ using a fixed effect model, after which we pooled single estimates using a random-effects model.

We performed subgroup analyses according to BMI (<25, 25-30, >30), smoking status (never, former, current) use of hormonal replacement therapy (never, ever), caffeination status (caffeinated,

decaffeinated), study location (Europe, North America), postmenopausal status for cohort studies, and the type of controls (hospital-based, population-based), postmenopausal status, and study location (Europe, Asia, North America), for case-control studies.

(29)

A dose-response analysis was conducted using generalized least squares for trend estimation method (125, 126). First, as the dosage values, we used medians reported in the studies or estimated them if the information was not available. Then, a variance-covariance matrix of the beta coefficients (log transformed RRs/ORs) in each of the studies was estimated to obtain dose-response relationship curves (125, 127).

We modeled restricted cubic splines with three knots positioned at 2%, 25%, and 80% percentile of the coffee consumption distribution for cohort studies, and at 5%, 45%, and 80% percentile of the distribution for case-control studies. A non-linear relationship was assessed by testing the null hypothesis that the coefficients of the second spline were equal to zero (128).

In addition to a chi-square test, the heterogeneity between the studies was quantified by I² statistic:

𝐼²= (𝑄 − df

𝑄 ) 𝑥 100%

where Q represents the value of chi squared statistics, i.e. the test that assesses if the observed differences in individual study results are compatible with chance alone, and df represents the tests degrees of freedom (129). The values ranging from 0-25% were considered as a low heterogeneity, from 26-50% as a moderate, and above 50% as a substantial heterogeneity between the studies (129).

The difference in meta-estimates between study designs and between subgroups were tested by fitting meta-regression models (method of moments) and by conducting Wald’s tests.

The risk of publication bias was tested by visual inspection of funnel plots presenting log-transformed effect estimates and its standard errors, and with Egger’s, and Begg’s tests at 10% significance level (130).

(30)

4 Results

4.1 Paper 1

During an average of 13.1 years of follow-up of 91 767 women, 9675 cases of cancer were diagnosed:

3277 (33.9%) breast cancers, 1266 (13.1%) colorectal cancers, 446 (4.6%) ovarian cancers, and 819 (8.5%) lung cancers.

Compared to light coffee consumers (≤1 cups/day), women who drank more than 3 and up to 7 cups/day had a 17% reduced risk of colorectal cancer (95% CI 0.70-0.98, ptrend across categories=0.10) and a 9% reduced risk of cancer at any site (95% CI 0.86-0.97, ptrend across categories=0.03). A significantly increased risk of lung cancer was observed in women who were heavy coffee consumers (>7

cups/day) compared to light coffee consumers (HR=2.01, 95% CI 1.47-2.75, ptrend across categories<0.001).

However, no statistically significant association was observed in never smokers (>5 cups/day vs. ≤1 cups/day HR=1.42, 95% CI 0.44-4.57, ptrend across categories=0.30). No significant association was found between coffee consumption and the risk of breast or ovarian cancer.

The results from the complete-case analyses were similar to those from the analyses of multiple imputed datasets. We did not find evidence of violation of proportional hazards assumption. None of the interactions tested reached statistical significance. After we excluded breast cancer cases

diagnosed during the first two years of follow-up, we found significantly decreased risk of breast cancer for low and high moderate coffee consumers compared to light consumers (HR=0.90, 95% CI 0.81-0.99; HR=0.86, 95% CI 0.78-0.96, ptrend across categories=0.01).

4.2 Paper 2

The average follow-up of 145 247 women and 48 192 men was 13.6 years of follow-up. During more than 2.6 million person-years, a total of 19 733 cancer cases were identified, out of which 479 (2.4%) were bladder cancers, 97 (0.5%) esophageal cancers, 475 (2.5%) kidney cancers, 491 (2.5%)

pancreatic cancers, and 281 (1.4%) stomach cancers.

Compared to light filtered coffee consumers (≤1 cup/day), heavy filtered coffee consumers (≥4 cups/day) had 26% reduced risk of being diagnosed with pancreatic cancer (95% CI 5%-43%, ptrend across categories=0.01). We did not observe significant associations between total or boiled coffee

consumption and any of the investigated cancer sites, neither in the entire study sample nor in analyses stratified by sex. The reduced risk of pancreatic cancer was confined to never smokers (moderate vs.

light filtered coffee consumers HR=0.60, 95% CI 0.41-0.87; heavy vs. light filtered coffee consumers HR=0.60, 95% CI 0.35-1.01; ptrend across categories=0.01).

(31)

Women who drank more than 1 and up to 4 cups of boiled coffee per day had a HR of 1.58 (95% CI 1.03-2.05) compared to light boiled coffee consumers. However, no association was found for women who drank more than 4 cups of boiled coffee, nor we found a statistically significant dose-response relationship. We found an increased risk of bladder cancer among never smokers who were heavy filtered coffee (HR=1.87, 95% CI 1.01-3.45, ptrend across categories=0.03) or heavy total coffee consumers (HR=1.86, 95% CI 1.12-3.10, ptrend across categories=0.08), and an increased risk of stomach cancer in never smokers who were heavy boiled coffee consumers (HR=2.14, 95% CI 1.10-4.16, ptrend across

categories=0.04).

No significant departure from linearity between filtered, boiled, and total intake and the study outcomes was found. The test of Schoenfeld residuals did not indicate a violation of the proportional hazards assumption. We found no evidence of effect modification between coffee consumption and BMI for any of the outcomes. The estimates for stomach cancer became stronger in the analyses performed after we excluded cases diagnosed during the first year of follow-up.

4.3 Paper 3

Twelve cohort (131-142) and eight case-control (143-150) studies were eligible for inclusion, contributing with 11 663 and 2 746 endometrial cancer cases, respectively. The summary RR for highest compared with lowest coffee intake category was 0.73 (95% CI: 0.67–0.80; pheterogeneity = 0.051, I² = 36.7%) after combining results from the cohort and case control studies. The corresponding summary RR among cohort studies was 0.76 (95% CI: 0.69–0.83; pheterogeneity = 0.06, I² = 40.5%) and 0.63 (95% CI: 0.53–0.76; pheterogeneity = 0.57, I² = 0%) for case-control studies. One-cup increment per day was associated with 3% risk reduction (95% CI: 2% -4%) in cohort studies and 12% (95% CI: 5%

-18%) in case-control studies. We did not observe a statistically significant difference between the estimate from cohort and case-control studies (p=0.1), after applying a meta-regression model. We found evidence of an under-representation of smaller studies after a visual inspection of a funnel plot.

After pooling the results from five cohort studies, the association remained significant only in women with body mass index over 30 (RR=0.71, 95% CI: 0.61-0.81). The summary RR for high vs low coffee consumption categories in postmenopausal women was 0.71 (95% CI: 0.62-0.81, pheterogeneity

= 0.394, I² = 4.2%). In cohort studies, a significant protective RR was observed for ever smokers (RR

= 0.67, 95% CI: 0.49-0.92; pheterogeneity = 0.005, I² = 73.5%), never smokers (RR = 0.78, 95% CI: 0.67- 0.92; pheterogeneity = 0.302, I² = 17.2%) and former smokers (RR = 0.80, 95% CI: 0.66-0.97; pheterogeneity

= 0.332, I² = 0%).

Inverse associations to risk of endometrial cancer were observed in sub-analyses of cohort studies for caffeinated (RR= 0.64, 95% CI: 0.49-0.83; pheterogeneity = 0.088, I² = 50.8%), decaffeinated coffee (RR=

(32)

therapy (RR=0.59, 95% CI: 0.49-0.73; pheterogeneity = 0.579, I² = 0%). A negative association between high coffee intake and endometrial cancer risk was further observed across different geographical regions (Europe, Asia, North America). The association between coffee consumption and EC risk was attenuated when we conducted meta-analyses using the alternative approach in which we compared moderate/heavy vs never/low coffee consumers.

(33)

5 Discussion

5.1 Summary of the results

In paper 1, a decreased risk of cancer at any site was associated with high moderate coffee

consumption. We found a statistically significant association between high coffee consumption (>7 cups/day) and increased risk of lung cancer. However, residual confounding due to smoking may have contributed to the association between high coffee consumption and the risk of lung cancer.

The data from the joint NOWAC and NSHDS cohort suggest that high filtered coffee consumption might reduce the risk of pancreatic cancer. No evidence of an association was found between coffee consumption and the risk of esophageal or kidney cancer. The positive association between bladder cancer risk and boiled coffee consumption in women who were moderate boiled coffee consumers was not supported by the findings from the dose-response analysis. Further, the increased risk of bladder and stomach cancer was found in never smokers.

The results from the meta-analysis strengthen the evidence of a protective effect of coffee consumption on the risk of endometrial cancer and suggest that the effect of coffee intake is particularly beneficial for women with obesity.

5.2 Discussion of methodology

5.2.1 Methodological considerations in prospective cohort design

5.2.1.1 Selection bias

Selection bias occurs when a sample of study participants is not representative of the source

population from which the sample was derived from, i.e. if there are systematic differences between those that participate in a study and those who do not (151). Several concerns can be raised in case of high percentage of non-responders. For example, if there are differences between responders

compared to non-responders when it comes to demographics, life-style, health or diet.

The response rate of 52.7% in the NOWAC cohort and the participation rate of 48-79% in the subcohorts of the NSHDS study are similar to those in other population-based cohorts (152). The higher response in the NOWAC cohort was observed in the age-groups 30–34 years till 55–59 years (60%), while 44.7% responded among those aged 65–70 years (115, 152). The authors of a validation study of the cohort found no differences in oral contraceptive use, parity, and the level of education between the responders and the non-responders. Moreover, the study participants did not differ from the

(34)

Similarly, in the VIP cohort, only limited evidence of selection bias related to income, age, and unemployment was found. In both genders, the youngest group was less represented among

participants. Participation was also lower in those that were unemployed compared to the employed, and in individuals that were in the lowest income level (153).

The quality assessment of the MONICA project revealed that compared to participants, non-

participants were more often regular cigarette smokers, had somewhat lower BMI, and that a smaller proportion were married or cohabitant. No substantial differences were found in level of education (154, 155). Overall, the authors of the quality assessment conclude that the data from the MONICA cohort is of good quality (154, 156).

Somewhat higher level of education observed in the NOWAC cohort, and the lower participation rates of those with a low income in the VIP cohort might have resulted in the measure of association between coffee intake and the risk of cancer that is, to an unknown degree, different from the source population due to a underrepresentation of heavy coffee consumers. Highly educated women in the NOWAC are less likely to be heavy smokers compared to lower educated women. As there is an observed positive correlation between smoking exposure and coffee intake, i.e. heavy smokers are more likely to be a heavy coffee consumers, this indicates that highly educated women drink less coffee compared to women with fewer attained years of education. Similar conclusion might be drawn in regards to the VIP cohort, as smoking and other forms of tobacco use are much higher among those with lower income, the group that was less likely to participate in the survey (157). Finally, as the non- participants in the MONICA cohort were more likely to be active smokers, it might be speculated that heavy coffee consumers are slightly underrepresented in the NOWAC and the NSHDS cohorts.

5.2.1.2 Information bias

Information bias arises in an epidemiological study when systematic errors in measurements of exposure and/or outcome have occurred (158). Measurement errors of exposure can result in

misclassification which can be divided in two groups. A non-differential misclassification is the result of measurement error of an exposure that occurred among participants independently of the study outcome. Contrary, a differential misclassification occurs when the extent of measurement error is different for those study participants with a disease and those without (151).

As the data on coffee consumption and other covariates were collected from self-administered food- frequency questionnaires (FFQ), a misclassification of a certain degree is possible. However, the FFQs used in the NOWAC cohort were validated by 24-hours dietary recalls study. The results from the validation study showed a high validity of information on coffee consumption (Spearman’s correlation coefficient r=0.82) (159). Moreover, a physical activity scale used to measure physical activity in NOWAC women, self-reported history of diabetes mellitus, as well as anthropometric measures, all

(35)

were shown to be reliable based on the results from the validation studies (160-162). Validation of food-frequency questionnaire measurements in the Northern Sweden Health and Disease cohort has shown that consumption frequencies of coffee were similarly measured by FFQ and 24-hours dietary recall regardless of preparation method (Pearson’s product moment correlations ranging between 0.72 and 0.84) (163).

In paper 1, we decided to use follow-up information on coffee consumption and on smoking exposure as the main confounder in order to take into account any intrapersonal variations that might have occurred during a follow-up of twenty years and thus reduce the risk of misclassification bias.

In both NOWAC and NSHDS cohorts, we lacked information on certain types of coffee drinks such as espresso, cappuccino, café latte, macchiato, or decaffeinated coffee as these brewing methods were not asked about in the questionnaires. Even though a consumption of these coffee types was uncommon in Norway and Sweden at the time of data collection, it is still possible that total coffee consumption was underestimated for some of the participants, thus resulting in misclassification that is most likely to be non-differential. We also did not have information on instant coffee consumption from the NSHDS cohort. As instant coffee is prepared by freeze-drying filtered coffee, from a chemical point of view, this distinction is of less importance from a misclassification of brewing technique perspective.

However, due to a lack of information on instant coffee intake, the total coffee consumption in Swedish cohort was somewhat underestimated.

When self-reported, smoking exposure measures such as smoking status, number of cigarettes smoked per day, or duration of active smoking are particularly prone to measurement errors as smoking prevalence is often underestimated (164). Indeed, in the sample from paper 1, we observed 1 722 women that reported being ever smokers at baseline and never smokers at follow-up, and the misclassification of this sort is unlikely to have occurred only in follow-up data. The observed risk estimates might have been stronger for the outcomes that were inversely associated with coffee consumption if the true distribution of smoking status was available from our data, whereas in case lung cancer risk, the observed positive association would have been weaker. We did not update information on other covariates other than coffee consumption and smoking exposure variables in paper 1. As the number of missing data in all covariates on follow-up was around 30%, we lacked computational power to perform multiple imputation on all the variables at both baseline and the follow-up. However, we conducted a complete-case analyses for all of the outcomes in which we had updated information on all of the covariates included in the analyses. The estimates from these analyses did not differ from the main results. For similar reasons, we could not use the repeated information from the third round of questionnaires in the NOWAC study in paper 1, nor have we updated information on the main exposure or covariates in paper 2. Due to proportion of missing in

Coffee and cancer

Coffee and cancer

Coffee and cancer

Acknowledgments

Table of Contents

Summary

Sammendrag

List of papers

Abbreviations

1 Introduction

1.1 Coffee

1.1.1 Biological classification and history

1.1.2 Coffee constituents

1.1.3 Cancer incidence in Norway

1.1.4 Cancer incidence in Sweden

1.2 Coffee consumption and the risk of cancer – the story so far

2 Aims

3 Material and methods

3.1 Papers 1 and 2

3.1.1 The Norwegian Women and Cancer Study

3.1.2 Northern Sweden Health and Disease Study

3.1.3 Study population

3.1.4 Information on coffee consumption and covariates

3.1.5 Cancer incidence, death, and emigration

3.1.6 Statistical analysis

3.1.7 Multiple imputation

3.2 Paper 3

3.2.1 Search strategy, study selection, and data extraction

3.2.2 Statistical analysis

4 Results

4.1 Paper 1

4.2 Paper 2

4.3 Paper 3

5 Discussion

5.1 Summary of the results

5.2 Discussion of methodology

5.2.1 Methodological considerations in prospective cohort design